Search Tuning Reference
This document collects the primary “magic numbers” used in ranking and performance tuning. The defaults are chosen for high signal with minimal noise.
Field Boost Values
| Value | Field | Rationale |
18.0 | alias_boost | Highest priority; exact alias matches indicate strong relevance |
12.0 | name_boost | Model/source names are primary identifiers |
10.0 | nova_metric | KPI/metric names have high business value |
8.0 | nova_measures | Measure definitions support analyst workflows |
7.0 | nova_synonyms | Business term aliases improve discoverability |
6.0 | description | Descriptions provide context but may be verbose |
4.0 | column | Column names are often technical, lower priority |
3.0 | tag | Tags are useful but often inconsistent |
2.0 | path | File paths rarely match user intent |
1.5 | code | SQL code has high false-positive rate |
Hybrid Search Constants
| Value | Setting | Rationale |
60.0 | rrf_k | Standard RRF smoothing constant |
3 | overfetch_multiplier / DBT_NOVA_SEARCH_RRF_OVERFETCH | Fetch 3x results for better reranking |
200 | vector_top_k | Balance recall and latency |
20 | rerank_top_n | Cross-encoder is expensive; limit to top results |
Fuzzy Matching
| Value | Setting | Rationale |
0.75 | levenshtein_threshold | 75% similarity catches typos without false positives |
4 | fuzzy_min_length | Avoid fuzzy matching for very short tokens |
2 | fuzzy_max_distance | Allow up to 2 edits |
Scoring Adjustments
| Value | Setting | Rationale |
0.6 | DBT_NOVA_SEARCH_STAGING_DEBOOST_FACTOR | De-boost entities mapped to a staging layer (staging/stage/stg) |
1.15 | DBT_NOVA_SEARCH_MEASURE_MATCH_MULTIPLIER | Boost results whose meta.nova.measures align with query terms |
1.2 | DBT_NOVA_SEARCH_METRIC_MATCH_MULTIPLIER | Boost KPI/metric term matches |
1.2 | DBT_NOVA_SEARCH_SYNONYM_MATCH_MULTIPLIER | Boost business synonym matches |
1.08 | DBT_NOVA_SEARCH_CANONICAL_MATCH_MULTIPLIER | Baseline uplift for canonical entities |
1.35 | DBT_NOVA_SEARCH_CANONICAL_META_MATCH_MULTIPLIER | Extra uplift when canonical + meta terms both match |
2.5 | DBT_NOVA_SEARCH_CANONICAL_META_MATCH_BONUS | Additive bonus for strong canonical/meta alignment |
2.0 | DBT_NOVA_SEARCH_ENGINEER_EXACT_MATCH_MULTIPLIER | Engineer bias toward exact ID/name matches |
Analyst Semantic Multipliers
| Value | Setting |
1.09 | DBT_NOVA_SEARCH_ANALYST_METRIC_DEF_MULTIPLIER |
1.05 | DBT_NOVA_SEARCH_ANALYST_MEASURE_DEF_MULTIPLIER |
1.05 | DBT_NOVA_SEARCH_ANALYST_GRAIN_MULTIPLIER |
1.03 | DBT_NOVA_SEARCH_ANALYST_TIME_FIELD_MULTIPLIER |
1.03 | DBT_NOVA_SEARCH_ANALYST_DIM_OVERLAP_ONE_MULTIPLIER |
1.06 | DBT_NOVA_SEARCH_ANALYST_DIM_OVERLAP_TWO_MULTIPLIER |
1.09 | DBT_NOVA_SEARCH_ANALYST_DIM_OVERLAP_THREE_PLUS_MULTIPLIER |
0.96 | DBT_NOVA_SEARCH_ANALYST_MISSING_METRIC_MEASURE_MULTIPLIER |
0.97 | DBT_NOVA_SEARCH_ANALYST_MISSING_GRAIN_MULTIPLIER |
0.85 | DBT_NOVA_SEARCH_ANALYST_MIN_MULTIPLIER |
1.35 | DBT_NOVA_SEARCH_ANALYST_MAX_MULTIPLIER |